ClassificationMixin

class torchgeo.trainers.ClassificationMixin(*args, **kwargs)

Bases: LightningModule

Mix-in for classification-based tasks.

Added in version 0.9.

configure_losses()

Initialize the loss criterion.

configure_metrics()

Initialize the performance metrics.

Includes the following metrics:

• Accuracy: \(\frac{TP + TN}{P + N}\)

• Precision: \(\frac{TP}{TP + FP}\)

• Recall: \(\frac{TP}{P}\)

• F1Score: \(\frac{2 TP}{2 TP + FP + FN}\)

• JaccardIndex: \(\frac{TP}{TP + FN + FP}\)

See https://en.wikipedia.org/wiki/Evaluation_of_binary_classifiers for more details.

Higher values are better for all metrics. All metrics report multiple versions:

• Overall (micro): Sum statistics over all labels

• Average (macro): Calculate statistics for each label and average them

• Classwise (none): Calculates statistic for each label and applies no reduction

on_train_epoch_end()

Log train metrics.

on_validation_epoch_end()

Log validation metrics.

on_test_epoch_end()

Log test metrics.